research article qos analysis for ... -...

Research ArticleQoS Analysis for a Nonpreemptive Continuous Monitoring andEvent-Driven WSN Protocol in Mobile Environments

Israel Leyva-Mayorga,1 Mario E. Rivero-Angeles,2

Chadwick Carreto-Arellano,3 and Vicent Pla1

1 ITACA, Universitat Politecnica de Valencia, 46022 Valencia, Spain2Communication Networks Laboratory, CIC-IPN, 07738 Mexico City, DF, Mexico3SEPI ESCOM-IPN, 07738 Mexico City, DF, Mexico

Correspondence should be addressed to Israel Leyva-Mayorga; [email protected]

Received 4 September 2014; Revised 24 December 2014; Accepted 25 December 2014

Academic Editor: Bulent Tavli

Copyright © 2015 Israel Leyva-Mayorga et al. This is an open access article distributed under the Creative Commons AttributionLicense, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properlycited.

Evolution in wireless sensor networks (WSNs) has allowed the introduction of new applications with increased complexityregarding communication protocols, which have to ensure that certain QoS parameters are met. Specifically, mobile applicationsrequire the system to respond in a certain manner in order to adequately track the target object. Hybrid algorithms that performContinuous Monitoring (CntM) and Event-Driven (ED) duties have proven their ability to enhance performance in differentenvironments, where emergency alarms are required. In this paper, several types of environments are studied using mathematicalmodels and simulations, for evaluating the performance of WALTER, a priority-based nonpreemptive hybrid WSN protocol thataims to reduce delay and packet loss probability in time-critical packets. First, randomly distributed events are considered. Thisenvironment can be used to model a wide variety of physical phenomena, for which report delay and energy consumption areanalyzed by means of Markov models. Then, mobile-only environments are studied for object tracking purposes. Here, some ofthe parameters that determine the performance of the system are identified. Finally, an environment containing mobile objectsand randomly distributed events is considered. It is shown that by assigning high priority to time-critical packets, report delay isreduced and network performance is enhanced.

1. Introduction

Performance in wireless sensor networks (WSNs) protocolis typically determined by its energy consumption underdesired conditions, which in turn determines the networklife-time (period of time in which the network is consideredto be functional). This is due to the energy restriction pre-sented in the nodes (small devices with sensing and wirelesstransmission capabilities, which are battery supplied). How-ever, as power electronics evolve, other quality of service(QoS) parameters are gaining relevance, such as report delayand packet loss probability. These parameters determine theoverall performance of the network. When monitoring time-critical events, such as disaster management or object track-ing/detection applications, these parameters become muchmore relevant. If the network fails to deliver an alarm

message in time, the entire environment may be put at risk,resulting in important information ormonetary losses for theuser. Furthermore, nodes are now capable of allocatingseveral sensors, hence granting the ability to monitor a widevariety of physical factors, such as temperature, soil moisture,air composition, and/or movement simultaneously [1–3].This increases network versatility but also jeopardizes perfor-mance, due to the fact that different data packets require dif-ferent QoS parameters. This is specially true when both con-tinuous monitoring and event detection applications mustcoexist in the same network.

In the literature it is common to find wireless sensornetworks designed for either continuous monitoring (CntM)[4, 5] or event-driven detection (EDD) applications [1, 6, 7].EDD WSNs are designed in order to send data in a sporadicfashion whenever a group of nodes detect an event of

Hindawi Publishing CorporationInternational Journal of Distributed Sensor NetworksVolume 2015, Article ID 471307, 16 pageshttp://dx.doi.org/10.1155/2015/471307

2 International Journal of Distributed Sensor Networks

interest. Conversely, in CntM networks, all nodes in thesystem are set to transmit data periodically to the sink node;as such, the end user always has updated data from thesurveyed region. Applications where both CntM and EDDare required have been largely overlooked. In EDDWSNs [6],communications are only triggered by the occurrence of aprespecified type of event, which typically comprises a sensormeasurement that exceeds a determined threshold, whichoccurs due to unusual conditions inside the network area. Inthese environments, it is advisable that data packets con-taining specific events are transmitted faster than the rest ofthe packets. These event reports may contain, for example,alarm messages related to explosions, chemical leaks, floods,or earthquakes and therefore represent high priority events.On the other hand, other types of events, such as a smallincrease in some parameters, do not require immediate trans-mission since they may represent slightly unusual or long-term conditions such as rain occurrence, environmental tem-perature increase/decrease, or pressure changes. The studiedprotocol in our previous work [8] presents a cluster-basedarchitecture with EDD and CntM capabilities. As such, whileno event is detected, the nodes transmit data in a periodicalfashion to the sink node in order to examine the evolution ofcertain parameters and obtain tendencies. This is performedusing a TDMA-like protocol in a clustered architecture. Oncean event is detected, the nodes inside the event area shift tocontention mode in order to send the event-related infor-mation as soon as possible using a random access protocol.The main goal of this protocol is to transmit event dataefficiently while sending data periodically, so the networkadministrators are aware of the adequate operation of the net-work. By limiting the CntM transmissions, network lifetimeand energy consumption are enhanced. Furthermore, when amultievent environment is considered, setting a higher trans-mission probability to time-critical packets significantly,improves QoS. A previous study is now extended by propos-ing a simple Markovian model in order to study energy con-sumption and packet latency in a hybrid WSN (where bothCntM and EDD are considered). Also, a priority scheme thatallows the transmission of different event-related packets bymeans of a random access (RA) protocol is proposed andstudied. As such, the priority scheme uses different transmis-sion probabilities in order to guarantee a lower report latencyin packets from a more important event. Finally, the casewhere the events are triggered by a mobile target is studiedin the context of the hybrid priority-based protocol.

The rest of the paper is organized as follows: Section 2reviews the previous work related to priority-based andmobility-aware protocols in the context ofWSNs, followed bythe network model and assumptions described in Section 3.In Section 4, the proposed protocol is described, and inSection 5 it is mathematically analyzed using a discrete-timeMarkov chain. Section 6 presents some relevant resultsderived from simulations and the analytical model for severalscenarios, including randomly distributed events, mobileenvironments, and a multievent environment, where objecttracking and environmental monitoring are performed. Thepaper concludes with a summary of results and conclusions.

2. Related Work

Hybrid protocols, such as APTEEN [7], have been proposedfor the transmission of both CntM and EDD applications.However, only randomly generated events have been consid-ered for performance analysis purposes. In [9], a transportlayer protocol is developed, in which priority is based on theinformation content of each data packet. Whenever a nodereceives several packets from its neighbors, it rearranges thosepackets according to their priority level, thus sending the highpriority packet first. Note that the packets, in this case, aregenerated periodically and the rate is controlled by the sinknode, so no collisions can occur. Furthermore, each nodegenerates a specific type of data, which means that all datapackets transmitted by a specific source have the samepriority whether or not an important event is detected.

In [10], a priority handling scheme has been presented,where CHs (cluster heads) define time intervals in which onlypackets that contain a high priority label are able to performchannel reservation duties. In case no high priority transmis-sion request is sent, low priority packets are able to contendfor channel utilization. While this approach theoreticallyreduces report delay for important packets, it requires moreenergy per event transmission due to request-to-send (RTS)and clear-to-send (CTS) messages. Furthermore, if severalnodes require using the channel during a single high priorityphase, collisions are likely to occur. In PSED [11], several typesof events can occur within the area of interest and priority isassigned based on the potential damage that the event repre-sents. Whenever an event is detected, data packets are sent tothe CH; then, the CH sends request messages to the mobilenodes within its transmission range, which are guided (basedon each event priority) to the detection zone for aiding theevent sensing nodes to transmit their packets reliably. Inthis particular protocol, transmissions occurring between thecluster members and the CH do not consider event priority,and hence event reporting delay is not reduced during thisphase. Also, since packet transmission is carried out by themobile nodes, report delay is highly dependent on their dis-tance from the event and their speed. In this case, mobility inthe WSN is achieved by granting mobile capabilities to adetermined number of nodes (such as adding a vehicle con-trolled by the network administrator or the network itself).However, mobility in WSNs may be introduced by meansof several other methods, such as granting the nodes thecapacity of detecting mobile entities [3] (by including prox-imity sensors on the nodes or by attaching a radio frequency(RF) transmitter to the mobile entity of interest). In theformer, mobility may be handled by defining synchroniza-tion intervals in the time frame such as in [12, 13]. Thisapproach reduces synchronization time with the nearest CH.Conversely, little research has been conducted for QoS instatic networks in charge of monitoring mobile entities withtransmission-only capabilities. This type of networks may beused to report several parameters ranging from vital signs[14, 15] to alarm messages and, by representing independentmobile objects (MOs), their behavior is not controlled neitherby the network nor by the user.

International Journal of Distributed Sensor Networks 3

The main differences between the aforementioned workand the one presented in this paper are as follows: priorityis assigned with a similar purpose as in [10], which is meantto improve report delay on relevant event packets that maycontain time-critical data. However, priority is handled byassigning different transmission probabilities to event packetsin a RA protocol operating on a slotted channel. Thisapproach is intended to increase the frequency of high prior-ity transmission attempts (considering that collisions occurbetween high and low priority packets) rather than per-forming channel reservation duties (which require a higheramount of transmissions per reported packet). Furthermore,an environment where independent mobile objects are dis-tributed within the network and its presence generates highpriority packets in the sensor nodes is studied. By doing this,results related to operational parameters that enhance QoSunder these conditions are presented, results that have beenscarcely studied in previous work.

3. System Model

In this section, the main network parameters and assump-tions considered throughout the paper are described.

This work is centered on a cluster-based WSN protocolnamed WALTER (WSN ALTernating CntM/ED block pro-tocol for nonpreemptive event reporting) with continuousmonitoring (CntM) and event detection (EDD) capabilities.This protocol was previously presented in [8] and additionalfeatures have been included for the present study. In particu-lar, we consider a clustering protocol similar to LEACH [4],where a certain number of clusters are formed. Each of thoseclusters contains a node called cluster head (CH). This nodegathers the information of all cluster members (CMs) andsends it directly to the sink node. The role of nodes acting aseither CHs orCMs shifts constantly throughout the operationof the network. This is done in order to avoid fast batterydepletion of nodes acting as CHs, since their transmissionsrequire high power in order to reach the sink node. Once thecluster is formed, nodes send the collected information basedon either random access or continuous monitoring, depend-ing on the generated time schedule. During CntM, data isreported periodically to the sink node using a contention-freetime division multiple access protocol (TDMA). Since thenodes transmit data continuously, resources are not wastedas each time slot is used by a particular sensor node insidethe cluster. Conversely, in the proposed protocol, for both thecluster formation and the event reporting phases a randomaccess protocol based on the NP/CSMA protocol is used.Since these transmissions only occur at certain moments inthe operation of the system, it is not practical to preassignresources to specific nodes for these purposes. Furthermore,it is not possible to predict which sensors would be active atany given time, due to the randomnature of both, joint packettransmission and event reporting. Hence, the active nodescontend among each other in order to gain access to themedium. It is then essential to carefully select the parametersthat trigger the detection and transmission for the randomaccess protocol in order to maintain an acceptable operation

of the network.This is usually determined byQoS parameterssuch as energy consumption and report latency. Specifically,the transmission probability assignment for high priority andlow priority events is of particular interest and will be furtheranalyzed in this work.

The following assumptions and system parameters areconsidered.The total number of sensor nodes in the system is𝑁tot = 100. Sensor nodes are uniformly distributed in an areabetween (0, 0) and (100, 100) meters (i.e., square 100 × 100

area).The sink node is located outside the supervised area at

the coordinate (200, 0). Hence, transmissions to the sink noderepresent, at minimum, a 100-meter transmission; thus theyare highly energy consuming. EachCHuses a distinct CDMAcode to transmit the gathered data to the sink node. As such,no collisions among CHs are possible. Average report delay isan important QoS parameter and studies [9, 16] have foundthat multihop networks may suffer from increased reportlatency.Then, the basis for this study is a single-hop network.However, themean increase in report latency due tomultihoptransmissions is studied to validate this statement. All sensornodes have the same amount of initial energy. The networkoperates on a slotted channel, with each slot representingthe required time for a data packet to be transmitted (as thetransmission bit rate is 40 kbps and data packet length is 2 kb,the duration of each time slot is 0.05 s). Randomly distributedevents are generated with probability 𝜀 = 0.02 when no nodehas packets awaiting for transmission and probability 𝜀 = 0

when nodes are still attempting any scheduled or eventtransmission; that is, nodes have a data buffer that can onlyallocate one event report. As such, whenever there is anoverlap of events (multiple events simultaneously), the sensornodes only transmit information relative to the first sensedevent. This is a reasonable assumption since the occurrenceof multiple events is unlikely to happen. Additionally, the useof high capacity data buffers would increment the cost of thenetwork. Several methods for event generation are consid-ered in order to study their impact on network performance,and each will be further explained in the following sections.Since the proposed protocol is nonpreemptive, no event-related transmission is performed during CntM phase andno CntM information is sent during EDD phase. The sizeof the data packet 𝑙 (2 kbits) comprises the data payload, theidentification field, Id, and a 𝑡𝑦𝑝𝑒 field to specify the type ofpacket: event packet and CntM data packet.These packets aresent during the Steady Phase, where data collecting and trans-mitting are performed according to a predetermined sched-ule. The size of the control packet is considered to be 1 kbits,which comprises the same fields but with shorter payload.Control packets are sent during the Setup Phase, which con-sists in theCluster Formation and Schedule Broadcast Phases.The energy consumed to transmit a packet depends on thelength of the packet 𝑙, the path loss exponent Pl (its value isset depending on the selected channel model: Pl = 2 for thefree space model, and Pl = 4 for the multipath model [4]),and the distance between the transmitter and receiver nodes,𝑑, as in [4]. Specifically,

𝐸𝑡𝑥(𝑙, 𝑑) = 𝑙 × 𝐸elec + 𝑙 × 𝜖

𝑓𝑠× 𝑑

Pl, (1)


Table 1: Parameter setting.

Parameter Value𝜖𝑓𝑠

10 pJ/bit/m2

𝐸elec 50 nJ/bitIdle power 13.5mWSleep power 15𝜇WInitial energy per node 10 JTransmission bit rate 40 kbps

where 𝐸elec is the electronics energy and 𝜖𝑓𝑠

× 𝑑Pl is the

amplifier energy that depends on the required transmissiondistance. In order to reduce the processing load in the pro-cessor of the nodes and ensure data delivery, two power levelsare defined for data transmission. A high energy transmissionis performed every time a CH sends data to the sink node.For calculating the required power for transmission, the CHis assumed to be located at (0, 100), which is the farthercoordinate inside the network from the sink, so the energyused for CH transmission is themaximum energy required toreach the sink from any coordinate within the area of interest,and hence the required energy for a CH transmission is

𝐸𝑡𝑥𝑐ℎ

(𝑙, 𝑑) = 𝑙 × 𝐸elec + 𝑙 × 𝜖𝑓𝑠× (√2002 + 1002)

Pl. (2)

On the other hand, a low-energy transmission is performedwhenever a CM sends data to its CH. In this case, clusternodes are set to transmit their data to up to 𝑑 = 35m.Therefore, the energy required for every CM packet is

𝐸𝑡𝑥(𝑙, 𝑑) = 𝑙 × 𝐸elec + 𝑙 × 𝜖

𝑓𝑠× (35)

Pl. (3)

The energy to receive a packet depends on the time thecommunication circuits must be enabled; hence, as the trans-mission rate is set to be 40 kbps, the total time required fortransmission depends only on packet size, which gives

𝐸𝑟𝑥(𝑙) = 𝑙 × 𝐸elec. (4)

It is worth noting that communication circuits in CHs mustbe on at all times in order to be able to receive packets fromtheir CMs in every phase, so the consumed energy of thesenodes depends on the number of time slots in each round.Each CH dissipates energy in receiving and transmitting thesignals received from the cluster nodes.The steady state phaseis considered to be of 20 seconds. A GB (geometric backoff)policy is employed for collision handling, with parameters 𝜏

ℎ

for high priority and 𝜏𝑙for low priority events. When con-

sideringmobility, RandomDirectionMobilityModel (RDM)[17] is considered due to the fact that the mobile entities tendto get a better distribution compared to Random WaypointMobility Model. The rest of the parameters are listed inTable 1.

Simulations in this work are conducted until one nodein the network has completely depleted its energy (exceptwhen network lifetime is studied) and delay is computed asthe average time needed for the transmission of every datapacket generated by an event for every cluster. Figure 1

CMCH

Event

CM to CHCH to BS

BS

(100, 100)

(0, 0) (200, 0)

Figure 1: Basic system model for performance analysis.

presents the basic system considered in this work. Note that asingle-hop system is assumed. The rationale behind this is toavoid the use of a routing algorithm that affects the energyconsumption on the system. This is because a multihopsystemmust decide the appropriate routes to relay the packetsthrough the multiple available choices. As such, nodes in apreferred route would consume more energy than nodes fur-ther away form the sink node. Since the objective of this workis to study the impact of the random access protocols throughthe use of priorities, we avoid the aforementioned effects ofthe selected routing protocol. Specifically, it would be verydifficult to know if a certain node consumed its energy due tothe use of the priority scheme or due to the fact that it was in ahigh traffic route in the network. Finally, the use of multihopschemes entails a higher packet delay compared to single-hop schemes. This issue will be explained in detail in a futuresection. Additionally, many routing protocols have beenproposed in the literature specifically aimed atWSNs applica-tions and the selection of one of these is not a trivial task. Assuch, we leave this issue for future work since it is outside thescope of this paper.

4. WALTER (WSN AlternatingCntM/ED Block Protocol forNonpreemptive Event Reporting)

Figure 2 presents the basic operation of WALTER, a hybridprotocol with CntM and EDD capabilities.

First, clusters are formed in the CF (cluster formation)interval according to [4]. Once the CF is over (when all nodesare either CMs or CHs and all nodes belong to a particularcluster), event detection is performed using a random access(RA) protocol, and it is denoted by the ED period. Wheneveran event packet is received by the CH, it is transmitted to thesink node using a particular CDMA code during the next


CF RA TDMA CDMA CF RA TDMA CDMA

ED CntM CH to BS CH to BSED CntM

Figure 2: Time frame for WALTER.

0

20

40

60

80

100

Activ

e nod

es

LEACHTEEN

WALTER

0 1000 2000 3000 4000 5000 6000 7000 8000Time (s)

(a) Pl = 2

0

20

40

60

80

100

Activ

e nod

es

0 200 400 600 800 1000 1200Time (s)

LEACHTEEN

WALTER

(b) Pl = 4

Figure 3: Network lifetime comparison between two characteristic WSN protocols and WALTER.

time slot.This is in order to avoid collisions between clusters.When the scheduled time for random access is over andno more cluster members have event data for transmission,TDMA is performed. It is worth noting that TDMA startsonly if every nodewithin a cluster has no event packets left fortransmission; in case that some event packets are awaitingtransmission, TDMA is delayed. This approach is usedbecause we consider that, in a hybrid protocol, event report-ing has higher priority than CntM data. At the end of theTDMA interval, the CHs transmit the gathered informationto the sink node.This process is repeated for 20 seconds,whenthe scheduled time for the next CF phase is reached.

The benefits of using a hybrid protocol capable of thetransmission of event-related and CntM data, as opposed tousing either an EDD or CntM specific protocol, have beenstudied in [8]. Particularly, a hybrid protocol presents a largernetwork lifetime than LEACHandTEENwhen operating in alow-rate event detection environment and when a CntM-likebehavior is needed.Network lifetime for an environmentwithrandomly generated events for LEACH, TEEN, andWALTERwith disabled priority is shown in Figure 3.

Figure 3 also illustrates the effects that different valuesof Pl have on the performance of the system. Specifically,the values of Pl = 2 and Pl = 4 are considered. Note thatconsidering a path loss exponent of Pl = 4 results in animmense increase in the required energy for the transmissionof a packet compared to the case where Pl = 2. Indeed, a highpath loss exponent is closer to the channel conditions foundin many practical environments. However, both simulationsand numerical methods used to solve the analytical modelstake considerably more time to perform. As such, the rest ofthe paper considers a value of Pl = 2. Also consider that

the general behavior of the studied protocols is the same forboth path loss values. For instance, note that in both casesWALTER outperforms LEACH and TEEN, and the maindifference between both models is the extra energy con-sumption in the system. Finally, a low value of the path lossexponent is closely related to open and obstacle free scenar-ios, such as the ones considered for mobile detection applica-tions, which is the focus of this work. The study of WALTERis now extended, as we now consider the case when thisprotocol is capable of detecting events with different prioritylevels. Event detection is performed using a double slidingwindow scheme [18] in order to limit transmissions generatedby each event. This implies that transmissions are triggeredonly at the first slot in which an event occurs (in whicha sensor lecture increases abruptly or surpasses a specificthreshold defined by the user, which may be related to theaccuracy level in the specific application). Hence, for a newevent-related transmission in a specific node to be triggered,theremust be at least a time slot in which no event is detected.Once an event has been detected, priority is assigned using adetection scheme where two threshold values are selected. Incase only the first threshold is exceeded, the detected event islabeled as low priority. However, if both thresholds areexceeded, the event is considered to be of high priority.Thesethresholds can be established at any time by the network oper-ator according to its particular interests.The rationale behindthis is as follows. Typically, whenever an event occurs, nodescloser to the coordinates of the origin of the event present ahigher value in their sensor lectures and have more detailedinformation about the event. As such, it is assumed that thisdata should be transmitted as soon as possible to the sink.On the other hand, nodes that still detect the event but with a


0

1

N N− 1 · · ·

1 − 𝜏(1 − 𝜏)N−1

1 − 𝜏(1 − 𝜏)N−2

𝜏𝜏(1 − 𝜏)

N−1𝜏(1 − 𝜏)

N−2

Figure 4: Markov model for nonpriority event reporting.

lower intensity are usually farther from the origin of the eventand their data can be transmitted afterwards.

5. Analytical Model

Given the WALTER protocol, a Markov model [19] is used toanalyze energy consumption and report delay depending onthe average number of nodes attempting transmission. First,the case where no priorities are enabled is considered. Theanalysis that follows is the same for the CF or event reportingphases. In the former, we start with𝑁 nodes, each one trans-mitting with probability 𝜏

𝑐. In the latter, the corresponding

values are 𝑁𝑒and 𝜏𝑒. Denoting these parameters generically

by 𝑁 and 𝜏 (values for 𝜏 are selected before deploying thenetwork), let 𝑆

𝑛denote the number of sensors that transmit

when there are 𝑛 nodes with a pending event or jointpacket transmission. 𝑆

𝑛is a binomial random variable with

parameters 𝑛 (0 ≤ 𝑛 ≤ 𝑁) and 𝜏. Then, 𝑃(𝑆𝑛

= 𝑗) =

(𝑛

𝑗 ) 𝜏𝑗(1 − 𝜏)

𝑛−𝑗 and 𝐸(𝑆𝑛) = 𝑛𝜏.

The aforementioned system can bemodeled as a discrete-time Markov chain,𝑊, depicted in Figure 4, where the statesrepresent the number of nodes that have not yet successfullytransmitted their packet.

The state space of 𝑊 is thus {𝑁,𝑁 − 1, . . . , 1, 0}, with𝑊(0) = 𝑁. Denoting𝑝

𝑛= 𝑃(𝑆

𝑛= 1) = 𝑛𝜏(1−𝜏)

𝑛−1, for 𝑛 ≥ 1,the nonzero transition probabilities are 𝑃(𝑊(𝑡 + 1) = 𝑛 − 1 |

𝑊(𝑡) = 𝑛) = 𝑝𝑛and 𝑃(𝑊(𝑡 + 1) = 𝑛 | 𝑊(𝑡) = 𝑛) = 1 − 𝑝

𝑛. To

these we add 𝑃(𝑊(𝑡 + 1) = 0 | 𝑊(𝑡) = 0) = 1 (i.e., 0 is anabsorbing state).

The time (number of time slots, each lasting 0.05 s) that𝑊 spends in state 𝑛, 𝑥

𝑠, is geometrically distributed: for any

state 𝑛 ≥ 1 and for 𝑚 ≥ 1, 𝑃(𝑇𝑛= 𝑚) = (1 − 𝑝

𝑛)𝑚−1

𝑝𝑛. The

mean time that the system remains in state 𝑛 is thus

𝐸 (𝑇𝑛) =

1

𝑝𝑛

. (5)

Therefore, the mean time to form the cluster is

𝐸 (𝑇cluster) =𝑁

∑

𝑛=1

[𝑛𝜏𝑐(1 − 𝜏

𝑐)𝑛−1

]−1

. (6)

On the other hand, the average event reporting time of all thenodes that sense the event is

𝐸 (𝑇event) =

𝑁𝑒

∑

𝑛=1

[𝑛𝜏𝑒(1 − 𝜏

𝑒)𝑛−1

]−1

. (7)

Finally, the average energy required for the transition ofthe system from state 𝑛 to 𝑛 − 1, 𝐸(𝐸

𝑛), is computed based on

the energetic cost of a transmission attempt, 𝐸𝑡𝑥, the required

energy for receiving a packet, 𝐸𝑟𝑥, and 𝜏. As stated above, 𝑥

𝑠

is the number of time slots required for transition. Finally, 𝛾is the energetic cost of a failed attempt. Then,

𝐸 (𝐸𝑛) =

∞

∑

𝑗=1

𝐸[𝐸𝑛

𝑥𝑠

= 𝑗]𝑃 (𝑥𝑠= 𝑗) , (8)

where

𝐸[𝐸𝑛

𝑥𝑠

= 𝑗] = 𝛾 (𝑗 − 1) + [𝐸𝑡𝑥+ (𝑛 − 1) 𝐸

𝑟𝑥] , (9)

which gives

𝐸 (𝐸𝑛) = 𝛾 (

1 − 𝑝𝑛

𝑝𝑛

) + 𝐸𝑡𝑥+ (𝑛 − 1) 𝐸

𝑟𝑥, (10)

where the energetic cost of a failed transmission attempt foreach time slot, 𝛾, is

𝛾 = (𝐸𝑡𝑥− 𝐸𝑟𝑥) (

𝑛𝜏 − 𝑝𝑛

1 − 𝑝𝑛

) + 𝑛𝐸𝑟𝑥. (11)

By substituting (11) in (10) we obtain the formula for therequired energy for the transition from state 𝑛 to 𝑛 − 1 asfollows:

𝐸 (𝐸𝑛) =

𝑛 ((𝐸𝑡𝑥− 𝐸𝑟𝑥) 𝜏 + 𝐸

𝑟𝑥)

𝑝𝑛

=(𝐸𝑡𝑥− 𝐸𝑟𝑥) 𝜏 + 𝐸

𝑟𝑥

𝜏 (1 − 𝜏)𝑛−1

.

(12)

This model may be used to calculate energy consumptionfor both event transmission and cluster formation by addingthe consumed energy for each 𝑛. As the process of clusterformation requires only one transmission from the CH, thetotal energy required to form a cluster is

𝐸 (𝐸cluster) = 𝐸𝑡𝑥𝑐ℎ

+

𝑁

∑

𝑛=1

(𝐸𝑡𝑥− 𝐸𝑟𝑥) 𝜏𝑐+ 𝐸𝑟𝑥

𝜏𝑐(1 − 𝜏

𝑐)𝑛−1

. (13)

For event reporting, the CH is set to transmit to the BS(base station) each time a packet containing event data isreceived; therefore,

𝐸 (𝐸event) = 𝑁𝐸𝑡𝑥𝑐ℎ

+

𝑁𝑒

∑

𝑛=1

(𝐸𝑡𝑥− 𝐸𝑟𝑥) 𝜏𝑒+ 𝐸𝑟𝑥

𝜏𝑒(1 − 𝜏

𝑒)𝑛−1

. (14)

As stated in Section 3, a single-hop network is considered.This is due to the fact that multihop delivery has been foundto present a higher latency when compared to single-hop[9, 20]. To evaluate this statement, consider the followingassumptions.

(i) While on TDMACHs are unable to send data directlyto sink, thus assuming that these nodes are able toreceive a relay request in these phases, they must waitfor RA to send the requested message.The number ofTDMA slots left is denoted by 𝑡𝑠.


P(S)

P(𝜏l)

P(𝜏l)

P(𝜏h) P(𝜏h) P(𝜏h)

N,M

N,M− 1

N, 0

N − 1,M

N− 1,M − 1

N − 1, 0

0,M − 1

0, 0

0,M· · ·

· · ·

· · ·

···

···

···

Figure 5: Markov model for high and low priority event reporting.

(ii) During steady state phases, a same number of timeslots are used for both TDMA and RA.Therefore, theprobability of requesting a packet retransmission by aspecific CH while being busy (performing TDMA) is𝑃(TDMA) = 1/2.

(iii) The desired number of clusters is set to 5; then, as𝑁tot = 100, each cluster has a mean number of CMs,𝑁cm = 19.Then, the average length of TDMA and RAphases is 𝐸(𝑇TDMA) = 𝑁cm + 1 = 20.

(iv) As the clusters operate independently, the probabilityof receiving a relay request in each time slot isuniformly distributed. Then, the average increase inreport delay due to a 𝑛-hop transmission is

𝐸 (Δ𝑇𝑛ℎ) = 𝑛ℎ × [

[

𝑃 (TDMA) ×𝑁cm

∑

𝑡𝑠=1

𝑡𝑠

𝐸 (𝑇TDMA)+ 1]

]

= 𝑛ℎ × [

𝑃 (TDMA)𝑁cm (𝑁cm + 1)

2𝐸 (𝑇TDMA)+ 1]

= 𝑛ℎ × [𝑃 (TDMA)𝑁cm

2+ 1] =

23

4𝑛ℎ.

(15)

Here, 𝑛ℎ is the number of hops required to reach the sinknode. Since we assume that there are packets with highpriority to be delivered, the use of a multihop scheme is notsuitable for the specific applications considered in this work.

A bidimensional Markov model as the one shown inFigure 5 is used for the case when nodes are able to detecthigh and low priority events. Here,𝑁 represents the numberof nodes detecting a high priority event and𝑀 represents thenumber of nodes that detect the low priority event. Recall thatnodes that detect a high priority event transmit with proba-bility 𝜏

ℎ, while nodes that detect a low priority event transmit

with probability 𝜏𝑙. Building on this, the system begins

at state 𝑊(0) = (𝑁,𝑀) and the state (0, 0) is an absorbingstate.The nonzero transitions probabilities at state (𝑖, 𝑗) are asfollows:

(i) to state (𝑖 − 1, 𝑗) with probability 𝑃ℎ

𝑠(𝑖, 𝑗) = 𝑖𝜏

ℎ(1 −

𝜏ℎ)𝑖−1

(1 − 𝜏𝑙)𝑗,

1 2 3 4 5 6 7 8Detections per cluster

FixedAverage

Mathematical

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Aver

age d

elay

(s)

Figure 6: Average event transmission delay in a nonpriorityenvironment for different values of reporting nodes per cluster.

(ii) to state (𝑖, 𝑗 − 1) with probability 𝑃𝑙

𝑠(𝑖, 𝑗) = (1 −

𝜏ℎ)𝑖𝑗𝜏𝑙(1 − 𝜏

𝑙)𝑗−1,

(iii) to state (𝑖, 𝑗) with probability 1 − 𝑃ℎ

𝑠(𝑖, 𝑗) − 𝑃

𝑙

𝑠(𝑖, 𝑗).

Let 𝑉𝑖,𝑗represent the average time it takes for the chain to go

form state (𝑖, 𝑗) to state (0, 0). This can be calculated throughthe following relationship:

𝑉𝑖,𝑗[𝑃ℎ

𝑠(𝑖, 𝑗) + 𝑃

𝑙

𝑠(𝑖, 𝑗)] − 𝑉

𝑖−1,𝑗𝑃ℎ

𝑠(𝑖, 𝑗) − 𝑉

𝑖,𝑗−1𝑃𝑙

𝑠(𝑖, 𝑗) = 1.

(16)

Specifically, we are interested in finding the average time thatboth the nodes that sense the high and low priority eventsreport their data. Hence, we numerically solve (16) for 𝑖 = 𝑁

and 𝑗 = 𝑀.Parameters section at the end of the paper depicts the

variables used for analyzing WALTER by means of thedescribed Markov model.

6. Numerical Results

6.1. Static Environments: Priority Disabled. QoS analysis isconducted using the Markov model described in Section 5and a discrete-time event simulator developed in C++. Theproposed Markov model is first used to validate simulationswhen no priorities are assigned; hence the transmissionprobability per time slot in each node during RA phases is𝜏 = 0.25.

Figure 6 shows the average delay for event packets withdifferent values of detections per event in each cluster. In thiscase, threeways for achieving event detections for each clusterwere considered, as shown in Figure 6: (a) a deterministicapproach, called Fixed strategy, where a certain number ofnodes per cluster are set to perform event detection duties,and then when every node in a cluster is affected by anevent, 𝜇 detections are triggered; (b) a probabilistic approach,called Average strategy, where a number of reporting nodesare chosen randomly and where a random number of nodesuniformly distributed in the ranges 0 and 2𝜇detect and reportthe event; and (c) a probabilistic approach, named Radius


strategy, where events are generated by means of a bidimen-sional random variable for the event center, (𝑋ev, 𝑌ev), and afixed detection radius, 𝑅.

In the Fixed strategy, it is assumed that the networkmanager decides the number of nodes that report their senseddata whenever an event occurs (given that the event affectsthe entire cluster); for example, 10% of the nodes in thecluster may report the event. This represents an efficientoption in order to limit the amount of transmissions in thenetwork, but the accuracy of the reported event may bereduced. The specific protocol to select these nodes is notstraightforward and we leave this issue for a future work.For the Average strategy, each CH broadcasts the value of auniformly distributed variable 𝑈(0, 2𝜇) during the setupphase, which determines the percentage of CMs that mayreport the event. Hence, the number of reporting nodes foreach Steady Phase may vary. Note that the mathematicalanalysis developed above closely matches the Fixed strategyresults. Indeed, for a deterministic number of nodes there isno variation in the number of reports per event. However, forthe Average and Radius strategies, the number of reportingnodes for each event is selected randomly. Hence, also thereporting delay varies. It is important to notice that this dif-ference ismore important for a high number of detections percluster, where there is a difference of approximately 16%between the mathematical and Average results.

By comparing the results in Figure 6 with (15), it maybe seen that the expected increase in report delay due tomultihop transmissions is significant. In fact, it is higher thanthe average delay for successfully reporting an event with asingle detection, even for 𝑛ℎ = 1. Then, despite the factthat (15) represents average delay increase for a single eventpacket, it illustrates that multihop is not a time-efficient strat-egy for critical-time applications. As a result of this, single-hop transmission is used for the rest of the paper.

Building on this, we study the effect of the event detectionradius. Here, uniformly distributed events are generatedwithin the area with center (𝑋ev, 𝑌ev) and radius 𝑅; that is, anevent occurring at a distance lower than or equal to𝑅 from thenode is detected. For this, the distribution of the simultaneousdetections for each event radius is calculated. From there,the average number of nodes detecting each event withina cluster is computed. Note that this approach is differentfrom theAverage strategy, where the average number of nodesreporting the event 𝜇 is determined a priori by the user.Figure 7 shows themean number of simultaneously triggereddetections within a cluster for different values of 𝑅.

In order to adequately calculate average report delay forrandomly distributed events, the proposed Markov modelmust include the probability of 𝑛 nodes detecting the event.Then, (7) becomes

𝐸 (𝑇event (𝑅)) =

𝑁𝑒

∑

𝑛=1

[𝑝𝑑(𝑛) × 𝑛𝜏

𝑒(1 − 𝜏

𝑒)𝑛−1

]−1

. (17)

By using (17), the error obtained between simulationand mathematical results is acceptable, as shown in Figure 8.However, computing the probability distribution of detecting

0

2

4

6

8

10

Aver

age d

etec

tions

per

clus

ter

0 5 10 15 20 25 30Event detection radius (m)

Figure 7: Average detection for each generated event in a nonpri-ority environment with uniformly distributed events with radius ofdetection 𝑅.

1 2 3 4 5 6 7 8Detections per cluster per event

MathematicalSimulation

0

1

2

3

4

Tim

e (s)

Figure 8: Average event report delay in a nonpriority environmentwith uniformly distributed events with radius of detection 𝑅.

nodes, 𝑝𝑑(𝑛) is not straightforward, specially for environ-

ments containing events with a higher complexity.Regarding the detection strategies, it can be seen that

the Fixed strategy achieves a lower reporting delay than theAverage and Radius strategies. As such, we focus on thisstrategy for the rest of the results presented in this section,as well as the basis for the mathematical model presented inthe previous section, when randomly distributed events areconsidered. As expected, the average delay increases alongwith the number of simultaneously triggered transmissionsper cluster. It is worth noting that, in order to achieve anadequate performance, a suitable threshold(s) value for eventdetection duties must be selected. Specifically, when set to ahigher value than needed, several events may be overlookedand, when set to a lower level than required, networkcongestion is likely to occur due to an increased rate of eventdetections and transmissions. Therefore, the network man-ager has to carefully select this threshold value.This is of spe-cial importance as energy consumption in contention modehighly depends on the number of simultaneously triggeredtransmissions for both cluster formation and event reportingphases. Figure 9 shows the total energy consumption perevent report for different values of nodes detecting the eventand a close match between the simulation and mathematicalresults is observed.


1 2 3 4 5 6 7 8 9 100

Detections per cluster per event

SimulationMathematical

0.02

0.04

0.06

0.08

0.1

0.12

0.14

Ener

gy (J

)

Figure 9: Average energy consumption per event transmission in anonpriority environment with uniformly distributed events.

5 10 15 20Event detection radius

GeneratedDetected

Transmitted

00.5

11.5

22.5

33.5

4

Even

ts

×104

Figure 10: Network activity for a nonpriority environment withuniformly distributed events with radius of detection 𝑅.

Specifically, the error between mathematical and simula-tion results is lower than 5% for all cases, which validates theenergy consumption model.

Also, this figure shows that energy consumption is nota linear function of the number of simultaneous detections.As such, the network user must be aware of the compromisebetween reliability and energy economy, achieved by select-ing the number of nodes per cluster that perform eventdetection duties, 𝜇.

When studying the Radius strategy, this behavior getsintensified. In Figure 10, several event detection radiuses areset by modifying the threshold value at the input of the sen-sors of the nodes; that is, for achieving a detection radius of5m, the threshold is set to a higher value and for the 20mradius it is set to a lower value.

From this figure it can be inferred that, by increasingthe threshold value, the detection radius decreases, so theprobability that an event is overlooked is higher. On the otherhand, as the threshold value gets lower, the event detectionradius increases along with the probability of generatingredundant data. Also, the number of detected and success-fully transmitted events is almost identical, which implies that

2 3 4 5 6 7 8 9 10Average speed (m/s)

1

0

2

3

4

5

Erro

r (%

)

Figure 11: Error percentage when comparing mathematical andsimulation results for the hitting time in a network with one mobileobject (MO) with detection radius𝑅 = 5m, 𝐿 = 52.14m,𝑇stop = 2 s,and𝑁 = 100.

packet loss probability is near to zero. Then, transmissionprobability is guaranteed, even when operating in environ-ments with high rate of event occurrence.

6.2. Mobile Environments: Priority Disabled. In case thatdetections within the area are generated by movement or thepresence of mobile objects (MO), a nonpriority environmentis considered. In this scenario, object position or status istransmitted by means of the RA protocol and environmentalparameters are reported during the collision-free TDMAschedule. By using (18) presented in [17], it is possible todetermine the expected hitting time (average time betweenmobile object detections within the network area) for theproposed system. Consider

𝐸𝑇𝑟𝑑

= (𝑁tot

2 ∗ 𝑅 ∗ 𝐿

)(𝐿

V+ 𝑇stop) . (18)

Here, 𝑁tot is the total number of nodes forming the net-work, 𝐿 is the mean length of the epoch, V is the average speedof the MO, and 𝑇stop is the mean time the MO remains staticfor eachmovement stage.While (18) was proposed for ad hocnetworks, the high count of nodes in the proposedWSN andtheir uniform distribution throughout the network present asimilar scenario to that of an ad hoc network.

For validation purposes, hitting time is obtained bysimulation and (18), the error obtainedwhen comparing bothresults, is more than acceptable as shown in Figure 11; then,both methods are considered suitable for calculating hittingtime for the selected environment.

Building on this, expected hitting time may be calculatedfor several values of the average speed of the MO, V, anddetection radius in the sensors by using (18). Results areshown in Figure 12.

As expected, hitting time decreases as the mean speedof the objects, V, or detection radius, 𝑅, increases. Forfurther research, these results may be used to estimate powerconsumption based on the average number of detections persecond, EDPs

𝑟𝑑= 1/𝐸𝑇

𝑟𝑑, and average energy consumed for

reporting an event with 1 simultaneous detection. It is worthnoting that (18) has been used when considering one mobile


2 4 6 8 10

Mobile detection radius (m)

05

10Average speed (m/s)

0

5

10

15

Expe

cted

hitt

ing

time (

s)

Figure 12: Expected hitting time for different V and 𝑅 obtained with(18).

2 3 4 5 6 7 8 9 10Mobile detection radius (m)

1 MO2 MO

4 MO

0

1

2

3

4

5

6

Expe

cted

hitt

ing

time (

s)

Figure 13: Expected hitting time when considering a number ofmobile objects (MO) and detection radiuses with V = 5m/s.

object within the area of interest. However, when severalmobile objects are included, its use may not be accurate. InFigure 13 several MOs are considered and a direct relationbetween hitting time for 1, 2, and 4 MOs is not observed.

This behavior occurs due to the fact that when a node hasdetected an event it is unable to perform a second detectionuntil the event packet has been successfully transmitted.Thus the rate of detection within the simulated networkwas decreased slightly, an scenario not considered by (18).Building on this, an equation for obtaining the expectedhitting time for 𝑛 mobile nodes is left for future work and,thus, results for environments considering several MOs areobtained by simulation.

Figure 14 shows that as detection radius increases, thedetections per second increase, which implies a proportionalincrease in energy consumption, given that the number ofsimultaneous detections per cluster remains constant. As theprobability of a node entering the detection radius of severalnodes in the exact same time slot is significantly low for thegiven scenario, detections are considered to be simultaneousif a detection is performedwhile any other node in the clusteris attempting the transmission of a previously detected MO.

Nevertheless, as Figure 15 shows, a significant increasein detections per cluster is not observed; in fact, as moremobile objects are added, this value decreases slightly. Thisoccurs due to the fact that as more objects transit the area,

2 3 4 5 6 7 8 9 100

0.5

1

1.5

22.5

3

3.5


Det

ectio

ns p

er se

cond

1 MO2 MO

4 MO

Figure 14: Detections per second when considering a number ofmobile objects (MO) and detection radiuses with V = 5m/s.

1

1.2

1.4

1.6

1.8

2 3 4 5 6 7 8 9 10Mobile detection radius (m)

1 MO2 MO

4 MO

Det

ectio

ns p

er cl

uste

r

Figure 15: Detections per cluster when considering a number ofmobile objects (MO) and detection radiuses with V = 5m/s.

the probability thatmore than oneMO iswithin the detectionrange of a node increases. But also, as MOs are added to thenetwork, detections are less spatially concentrated and morespread throughout the network.

It can also be seen from Figure 15 that, as the averagespeed of the MOs increases, the detections per clusterincrease. However, that increase is not significant, which alsoimplies that average report delay is not expected to increase.

This can be seen in Figure 16, where average report delayfor an environment containing a single MO is considered.

As stated above, adding MOs to the network does notincrease simultaneous detections, so average report delay isexpected to remain constant for these scenarios.While reportdelay is not significantly affected by neither detection radiusnor average speed nor the number of objects inside the areaof interest, other QoS parameters may be affected, such as theenergy consumption in the system.

Due to the fact that RA transmissions are high energyconsuming, an increase in the rate of detection of the eventswould imply an increase in power consumption; this behavioris shown in Figure 17.

It can also be observed that, in this case, energy consump-tion is mostly affected for scenarios with higher detections


2 3 4 5 6 7 8 9 10Average speed (m/s)

0.16

0.18

0.2

0.22

0.24

0.26

Aver

age r

epor

t del

ay (s

)

Figure 16: Average report delay for an environmentwith onemobileobject for several values of V.

2 3 4 5 6 7 8 9 100.01

0.015

0.02

0.025

0.03

0.035

0.04


Con

sum

ed p

ower

(J/s

)

1 MO2 MO

4 MO

Figure 17: Energy consumption for the Steady Phase when consid-ering a number of mobile objects (MO) and detection radiuses withV = 5m/s.

per second, rather than detections per cluster. Therefore, byincreasing the rate of detection, the consumed energy pertime unit is also increased. Building on this, the networkoperator has to achieve a balance between detection accuracyand low energy consumption, which may be controlled bycarefully selecting the threshold value for detection, which inturn leads to achieving the desired detection radius, 𝑅.

6.3. Static Environments: Priority Enabled. As reviewed pre-viously in this section, randomly distributed events that affectseveral nodes simultaneously tend to affect report delay andenergy consumption (as a result of increased rate of detec-tion/transmission) significantly. Hence, assigning priority tocertain packets is expected to enhance the performance of thesystem. In this work, priority is enabled by assigning differenttransmission probabilities, namely, 𝜏

ℎand 𝜏

𝑙, according to

the lectures in the sensor in charge of event detection. Thisis achieved by selecting two different threshold values. Forinstance, a higher transmission probability, 𝜏

ℎ, is selected

when both upper and lower thresholds are exceeded. Con-versely, a lower transmission probability, 𝜏

𝑙, is selected when

the lower threshold is exceeded but not the upper threshold.We now investigate the effect of the values of 𝜏

ℎand 𝜏𝑙on the

00.5

1

00.20.40.60.8

Aver

age d

elay

(s)

100

102

104

𝜏h

𝜏l

Figure 18: Average report delay for 𝑁 = 5 and 𝑀 = 5 for severaltransmission probabilities, obtained using the Markov model.

00.5

1

00.20.40.60.8D

elay

(s)

High priority

Low priority

10−1

102

101

100

𝜏h𝜏l

Figure 19: Delay improvement for high priority packets for severaltransmission probabilities.

performance of the system in order to find those that increaseperformance for the given environment.

In Figure 18, it can be seen that the lowest overall delay isachieved when selecting relatively low values of the transmis-sion probability: near 𝜏

ℎ= 0.3 and 𝜏

𝑙= 0.15. In this case, a

number of triggered high and low priority detections,𝑁 and𝑀, respectively, are set to be 𝑁 = 𝑀 = 5. This is in orderto recreate a highly collided environment so the system istested in the worst case scenario. Additionally, Figure 19shows the average report delay for high and low prioritypackets. Since the main goal of the priority scheme is to offera lower reporting delay to high priority packets, the valueswhere 𝜏

ℎ< 𝜏𝑙must not be considered and are shown to

demonstrate that only transmission probabilities are beingmodified. From this figure it is clear that, in order to achieve alower delay for high priority packets, the appropriate values ofthe transmission probabilities must be set to 𝜏

ℎ= 0.3 and 𝜏

𝑙=

0.15. As such, these values will be used as default throughoutthe rest of the paper.

In order to evaluate whether the selected priority han-dling scheme benefits the transmission of important packetsunder several conditions, average report delay for high andlow priority packets has been obtained for different values of𝑀 and𝑁 and is shown in Figure 20.

As expected, delay for high priority packets is significantlyimproved when compared to low priority packets. It is worthnoting that average report delay is computed as the time


12

34

5

12

34

50

1

2

NM

Del

ay (s

)

High priority

Low priority

Figure 20: Average report delay for high and low priority packetsfor different𝑁 and𝑀 using the Markov model.

1 2 3 4 5

12

34

50

5

10

15

NM

Erro

r (%

)

Figure 21: Error percentage for average report delay between theMarkov model and simulator.

it takes for the transmission of every packet correspondingto each priority level within a cluster, which is the reasonfor high priority packets experimenting higher delay when𝑁 > 𝑀+ 2. In this case, more high priority than low prioritypackets must be sent for the high priority event to besuccessfully transmitted. However, this scenario is difficult tooccur due to the inherent nature of high priority detections.Hence, the selected priority handling scheme performs asdesired. For validating the developeddiscrete event simulator,both analytical and simulation results regarding report delayhave been compared and the obtained error is presented inFigure 21. It can be seen that, in general, there is a goodmatch between the analytical model and simulation results.As the number of nodes in a cluster corresponds to a randomvariable, as 𝑁 + 𝑀 increases, so does the probability that acluster contains a lower number of CMs; then, when com-paring delay results for 𝑁 and 𝑀 detecting nodes (obtainedby means of the Markov model) and min(𝑁,CM) andmin(𝑀,CM−𝑁), results are likely to vary, which is the reasonfor the increase in the error for high values of 𝑁 and 𝑀. Inorder to adequately obtain average report delay in these cases,the probability distribution for high and low priority simul-taneous detections must be calculated and used in a similarmanner as in (17) for the two priority models. In order toverify that the selected priority scheme is economicallyadequate, energy consumption for high and low priority

0 0.2 0.4 0.600.10.20.30.40.50.6

0

0.5

1

1.5

Ener

gy co

nsum

ed (J

/eve

nt)

High priority

Low priority

𝜏h𝜏l

Figure 22: Energy consumed for several transmission probabilities.

2 4 6 8 10 12 14 16 18 200

0.2

0.4

0.6

0.8

1

1.2

1.4

Event detection radius (m)

Repo

rt d

elay

(s)

High priorityLow priority

Figure 23: Delay improvement for high priority packets in aconcentric detection scheme.

events is calculated for several values of the transmissionprobabilities, 𝜏

ℎand 𝜏𝑙, and shown in Figure 22.

It is observed that unless these values are set extremelyhigh, that is, 𝜏

ℎand 𝜏

𝑙above 0.5, which incurs in a highly

collided environment, consumed energy per transmission isnot significantly affected. Furthermore, the consumed energyby high and low priority transmissions is almost identical.Then, for achieving better performance for both, energyconsumption and report delay, lower values for transmissionprobabilities must be used, specifically, values near 𝜏

ℎ= 0.3

and 𝜏𝑙= 0.15 as suggested previously. Finally, results for the

case where randomly distributed events with detection radius𝑅 are considered are shown in Figure 23.

In this case, it is considered that nodes near the event cen-ter (selected by means of a bidimensional uniform randomvariable) are assigned high priority. Specifically, detectionswithin 𝑅/2 from the selected coordinate are set to behigh priority detections. On the other hand, detectionsbetween 𝑅/2 and 𝑅 are assigned a low priority label. Recallthat the rationale behind this priority scheme is that someenvironment monitoring applications [1] like wildfire [2] andmoisture or seismic monitoring may affect large areas. How-ever, their intensity fades as the distance from the initialoccurring site increases. As observed, by using the selected


2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8

1

Average speed (m/s)

Det

ectio

ns p

er se

cond

HP simulatedHP calculated

LP simulated

Figure 24: Detections per second for high and low priority packetsin a hybrid environment.

values for 𝜏ℎand 𝜏𝑙, delay in high priority packets is reduced

and shows better tolerance to situations where a high numberof transmissions are triggered simultaneously. In this case, itis assumed that every transmission related to an event detec-tion, regardless of the distance, is triggered simultaneously forevaluating the network in the worst case scenario. By doingthis, a high packet load environment is created, where highpriority packets compete directly with low priority ones forgaining channel usage. This behavior can be found in somespecific applications where explosions or chemical leaks canoccur.Whenusing the studied protocol formonitoring eventswith a low spreading rate, high priority packets are likely tobe the first ones to trigger transmission, thus leading to betterperformance.

6.4. Multievent Environments: Priority Enabled. Resultsshown in previous sections suggest that energy consumptionis significantly affected as detections per second increase.Thisis specially true when V, 𝑅, or even the number of mobileobjects (MOs)within the network increases.The following setof results shows the report delay for mobile object detectionswhen considering a hybrid environment containing mobileand randomly distributed events. Due to the importance ofaccurately reporting the detection or status of amobile object,packets containing this type of data are considered to be ofhigh priority. Therefore, randomly distributed events withdetection radius 𝑅 are considered to be of low priority. First,an environment where the detection radius for both lowpriority and the mobile object is fixed with 𝑅 = 5m isstudied.

Figure 24 shows the average expected (mathematical) andobtained (simulation) detections per second for high priorityevents. Also, low priority detections per second are shown. Asobserved, specially as average speed increases, low priorityevents interfere with the detection of the mobile object, lead-ing to loss of information. However, as Figure 25 shows, highpriority packets do not experiment a significant increase inreport delay, validating the proposed priority model.

Finally, an experiment is conducted by increasing detec-tion radius and defining V = 5m/s. In this case, the number of

0.17

0.175

0.18

0.185

0.19

0.195

2 3 4 5 6 7 8 9 10Average speed (m/s)

Aver

age d

elay

(s)

Figure 25: Average delay for high priority packets in a hybridenvironment.

2 3 4 5 6 7 8 9 100

0.2

0.4

0.6

0.8

1

1.21.4

Detection radius (m)

Det

ectio

ns p

er se

cond

HP simulatedHP calculated

LP simulated

Figure 26: Detections per second for high and low priority packetsin a hybrid environment.

low priority detections increases along with 𝑅, which createsa more hostile environment for mobile object detections.

This can be seen in Figure 26 where several mobiledetections are ignored due to the fact that the sensing node isin a busy state (where the node is trying to send informationrelated to a low priority detection).

Despite the fact that the number of detections increases,delay in high priority packets (Figure 27) tends to behave in asimilar manner compared to the previous experiment, wherelow priority detections were defined constant.

This is due to the fact that the probability for a mobileobject to be detected inside a busy section of the network, forexample, a cluster, is significantly low. Thus, report delay forhigh priority mobile events is nearly independent from theoccurrence of randomly distributed events.

On the other hand, it is likely that somemobile detectionsare ignored due to the fact that low priority events triggerthe transmission of several nodes and while transmissionattempts are performed, these nodes are considered unableto detect any other events. It is worth noting then that, in thisscenario, important data is lost.

7. Conclusions and Future Work

In this work, WALTER, a nonpreemptive hybrid protocol forWSN, is studied by means of a proposed Markov model


2 3 4 5 6 7 8 9 100.17

0.175

0.18

0.185

0.19

0.195

Detection radius (m)

Aver

age d

elay

(s)

Figure 27: Average delay for high priority packets in a hybridenvironment.

and a discrete event simulator. The aforementioned protocolimproves network lifetime and energy consumption for lowrate event detection environments. Throughout this study,WALTER performance is analyzed under several conditionsby defining a wide variety of environments. For each environ-ment, the proposedMarkov model may be used to accuratelyobtain the average report delay and the energy consumed perevent transmission. Some key aspects regarding the perfor-mance of WALTER are as follows.

(i) Fixed Event Detection Strategy. It is observed that averagereport delay increases in a nonlinear fashion along withthe number of simultaneous detections. By selecting theFixed strategy, average report delay and energy consumptionmay be calculated by means of a Markov model similarto the one presented. However, as 𝑁 and 𝑀 represent thenumber of CMs set to perform event detection duties, thecalculated QoS parameters correspond to scenarios in whichevents affect the clusters entirely. Whenever an event affectscertain section of the cluster, transmission is triggered in alower number of nodes. This causes event report delay andenergy consumption to adopt lower values than calculated.So, the presented performance analysis for the Fixed strategyrepresents the maximum values that can occur for averagereport delay and energy consumption. In case the networkuser requires to calculate the exact values for those param-eters, the described Average and Radius strategies may beused. However, it implies that the user must be aware ofthe probability distribution function of the detecting nodes,which is not straightforward to calculate. Furthermore, giventhe worst possible performance for a given scenario meetsthe monitoring needs of the user, performance is guaranteedthroughout the operation of the network.

(ii) Radius Event Detection Strategy. As mentioned in previ-ous sections, the studied environments such as concentricdetections are modeled in such a manner that when anevent occurs, every node involved in the detection processswitches to transmissionmode simultaneously.This behaviorcan de found in some military or industrial applications. Onthe other hand, when considering events that spread slowlywithin the network, high priority transmissions are expectedto occur before low priority detections, hence loweringcongestion within the clusters and enhancing performance.

(iii) Mobile Environments. When operating in this type ofenvironments, detections per second can be calculated inorder to predict whether an increase in energy consumptionis expected. During this study,𝐸𝑇

𝑟𝑑is used tomodel the aver-

age time needed for a single mobile object to be detected by astatic node, however this equation should not be used whentrying to determine the number of detections per secondwhen several mobile objects are considered.This is due to thefact that, for the given detection scheme, events are ignoredwhen several objects are within range of detection or thedetecting node has any RA transmission left.

(iv) Multievent Environments. The equation used to calculate𝐸𝑇𝑟𝑑can be used to obtain the percentage of ignored mobile

detections when multievent environments are considered.By calculating 1/𝐸𝑇

𝑟𝑑the average detections per second are

modeled. Then, this same parameter is calculated using thesimulator. By comparing these values the average missedmobile detections per minute may be obtained, whichrepresents an important QoS parameter for a particularapplication.

(v) Future Work. Despite the extensive analysis performedduring this study, some issues are left for future work,such as developing an algorithm for adequately selecting thereporting nodes in the Fixed strategy for enhancing energyconsumption and event report delay while maintaining eventreport probability. As stated above, (18) may be used toaccurately obtain hitting time in mobile-only environmentscontaining a single MO. A new equation may be proposed toadequately model this parameter in mobile environmentswith severalMOs or inmultievent environments. For this, theequationmust consider report delay and the probability of theMO entering to zones affected by other events. However, thisis not straightforward.

Parameters

𝑁: Number of nodes attemptingtransmission, high priority or prioritydisabled

𝑀: Number of nodes detecting a low priorityevent

𝜏: Transmission probability, priority disabled𝜏𝑐: Transmission probability during CFphases

𝜏𝑒: Transmission probability during eventreporting phases

𝜏ℎ: Transmission probability for high prioritypackets

𝜏𝑙: Transmission probability for low prioritypackets

𝑛: Nodes with a pending event or jointpacket transmission

𝑆𝑛: Nodes attempting transmission for a giventime slot

𝑊: State space for the proposed MarkovModel

𝑝𝑛: Successful transmission probability


𝑥𝑠: Number of time slots the system stays at a

given state𝑇cluster: Number of time slots required to form a

cluster𝑇event: Number of time slots required to send

every event packet𝐸𝑛: Mean energy required for transition in a

given state𝐸𝑡𝑥: Energy required for a transmission

attempt by a CM𝐸𝑡𝑥𝑐ℎ

: Energy required for a transmissionattempt by a CH

𝐸𝑟𝑥: Energy required for receiving a packet

𝛾: Energetic cost of a failed transmission𝐸(Δ𝑇𝑛ℎ): Mean increase in report delay for

multihop transmissions𝑛ℎ: Number of hops needed to reach the sink

node𝑃(TDMA): Probability that a cluster is performing

TDMA during steady state𝑁cm: Average number of CMs in each cluster𝑡𝑠: Remaining TDMA slots at a given time

slot.𝐸(𝑇TDMA): Mean number of time slots required for

TDMA.𝑃ℎ

𝑠(𝑖, 𝑗): Transition probability from state (𝑖, 𝑗) to

(𝑖 − 1, 𝑗)

𝑃𝑙

𝑠(𝑖, 𝑗): Transition probability from state (𝑖, 𝑗) to

(𝑖, 𝑗 − 1)

𝑉𝑖,𝑗: Mean time slots to absorption from state

(𝑖, 𝑗).

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper.

Acknowledgments

This work was partially supported by CONACyT underProject 183370.The research of Vicent Pla has been supportedin part by the Ministry of Economy and Competitiveness ofSpain under Grant TIN2013-47272-C2-1-R.

References

[1] T. Arampatzis, J. Lygeros, and S. Manesis, “A survey of applica-tions of wireless sensors and wireless sensor networks,” in Pro-ceedings of the IEEE International Symposium on MediterreanConference on Control and Automation, vol. 1, pp. 719–724,Limassol, Cyprus, June 2005.

[2] C. Ramachandran, S. Misra, andM. S. Obaidat, “A probabilisticzonal approach for swarm-inspired wildfire detection usingsensor networks,” International Journal of Communication Sys-tems, vol. 21, no. 10, pp. 1047–1073, 2008.

[3] S. Misra, S. Singh, M. Khatua, and M. S. Obaidat, “Extractingmobility pattern from target trajectory in wireless sensor net-works,” International Journal of Communication Systems, vol. 28,no. 2, pp. 213–230, 2015.

[4] W. B. Heinzelman, A. P. Chandrakasan, and H. Balakrishnan,“An application-specific protocol architecture for wirelessmicrosensor networks,” IEEE Transactions onWireless Commu-nications, vol. 1, no. 4, pp. 660–670, 2002.

[5] O. Younis and S. Fahamy, “Distributed clustering in Ad-Hocsensor networks: a hybrid, energy-efficient approach,” in Pro-ceedings of the 23th Annual Joint Conference of the IEEEComputer and Communications (INFOCOM ’04), IEEE, March2004.

[6] A.Manjeshwar andD. P.Agrawal, “TEEN: a routing protocol forenhanced efficiency in wireless sensor networks,” in Proceedingsof the 15th International Parallel and Distributed ProcessingSymposium, pp. 2009–2015, San Francisco, Calif, USA, April2000.

[7] A. Manjeshwar and D. P. Agrawal, “APTEEN: a hybrid protocolfor efficient routing and comprehensive information retrievalin wireless sensor networks,” in Proceedings of the InternationalParallel and Distributed Processing Symposium (IPDPS ’02), Ft.Lauderdale, Fla, USA, April 2002.

[8] I. L. Mayorga, M. E. Rivero-Angeles, C. C. Gutierrez-Torres, J.A. Jimenez-Bernal, R. Rodriguez, and A. Torres-Rivera, “Datatransmission strategies for event reporting and continuousmonitoring applications in wireless sensor networks,” in Pro-ceedings of the 7th International Conference on Broadband,Wireless Computing, Communication and Applications, pp. 1–7,November 2012.

[9] A. Sharif, V. Potdar, and A. J. D. Rathnayaka, “Prioritizing infor-mation for achievingQoS control inWSN,” in Proceedings of the24th IEEE International Conference on Advanced InformationNetworking and Applications (AINA ’10), pp. 835–842, IEEE,Perth, Australia, April 2010.

[10] V. J. Alappat, K. Nitish, and K. K. Anoop, “Advanced sensorMAC protocol to support applications having different prior-ity levels in wireless sensor networks,” in Proceedings of the6th International ICST Conference on Communications andNetworking in China (CHINACOM ’11), pp. 340–343, Harbin,China, August 2011.

[11] K. M. Alam, J. Kamruzzaman, G. Karmakar, and M. Murshed,“Priority sensitive event detection in hybrid wireless sensornetworks,” in Proceedings of the 21st International Conference onComputer Communications andNetworks (ICCCN ’12),Munich,Germany, August 2012.

[12] A. Raja and X. Su, “A mobility adaptive hybrid protocol forwireless sensor networks,” in Proceedings of the 5th IEEEConsumer Communications and Networking Conference (CCNC’08), pp. 692–696, IEEE, Las Vegas, Nev, USA, January 2008.

[13] B. Srikanth, M. Harish, and R. Bhattacharjee, “An energy effi-cient hybrid MAC protocol forWSN containing mobile nodes,”in Proceedings of the 8th International Conference on Infor-mation, Communications and Signal Processing (ICICS ’11),December 2011.

[14] Y.-D. Lee, D.-U. Jeong, and H.-J. Lee, “Empirical analysis of thereliability of low-rate wireless u-healthcare monitoring appli-cations,” International Journal of Communication Systems, vol.26, no. 4, pp. 505–514, 2013.

[15] K. S. Deepak and A. V. Babu, “Improving energy efficiencyof incremental relay based cooperative communications inwireless body area networks,” International Journal of Commu-nication Systems, vol. 28, no. 1, pp. 91–111, 2013.

[16] Y. Li, W. Ye, and J. Heidemann, “Energy and latency controlin low duty cycle MAC protocols,” in Proceedings of the IEEE


Wireless Communications and Networking Conference (WCNC’05), pp. 676–682, March 2005.

[17] T. Spyropoulos, K. Psounis, and C. S. Raghavendra, “Perfor-mance analysis of mobility-assisted routing,” in Proceedings ofthe 7th ACM International Symposium on Mobile Ad HocNetworking and Computing (MobiHoc ’06), pp. 49–60, Florence,Italy, May 2006.

[18] H. Shu and Q. Liang, “Fundamental performance analysis ofevent detection in wireless sensor networks,” in Proceedings ofthe IEEE Wireless Communications and Networking Conference(WCNC ’06), pp. 2187–2192, Las Vegas, Nev, USA, April 2006.

[19] G. Bianchi, “Performance analysis of the IEEE 802.11 distributedcoordination function,” IEEE International Journal on SelectedAreas in Communications, vol. 18, no. 3, pp. 535–547, 2002.

[20] W. Ye, J. Heidemann, and D. Estrin, “An energy-efficient MACprotocol for wireless sensor networks,” in Proceedings of the 21stAnnual Joint Conference of the IEEE Computer and Communi-cations Societies (INFOCOM ’02), pp. 1567–1576, June 2002.

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation http://www.hindawi.com

Journal ofEngineeringVolume 2014

Submit your manuscripts athttp://www.hindawi.com

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawi Publishing Corporation http://www.hindawi.com

Volume 2014

The Scientific World JournalHindawi Publishing Corporation http://www.hindawi.com Volume 2014

SensorsJournal of


Modelling & Simulation in EngineeringHindawi Publishing Corporation http://www.hindawi.com Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks


research article qos analysis for ... -...

Documents